Set up Case Management Agent to create and update cases

Case Management Agent streamlines the case management process, reducing manual effort and data entry errors.

You can use the creation and update feature of Case Management Agent to do the following actions:

• Create cases autonomously from conversations in voice, live chat, and other digital messaging channels. The agent identifies key details and creates cases automatically.
• Predict and update case fields autonomously in the following scenarios:
  • When the AI agent creates a case autonomously from a conversation
  • When a customer service representative (service representative or representative) manually creates a case from an email or conversation
  • When a case is created from an incoming email using automatic record creation and update rules

Prerequisites

• Administrators must have the CSR Manager or System Administrator role.

• Enable AI form fill assistance in the Power Platform admin center application.

• Automatically create or update records are set up to create case records from emails.

• The following configurations are set up if you want to create a case from conversations:

  • Provision channels in Dynamics 365 Contact Center.
  • Make sure you configure authenticated chat. Case Management Agent can create and update cases from authenticated chats only.
  • Create and manage workstreams and Create and manage queues for unified routing are set up.

• Move data across regions for Copilots and generative AI features in the Power Platform admin center application.

• Case Management Agent uses the Data Entry Agent in the background. The Power Platform Pay-as-you-go plan mandates the usage of an Azure subscription the system charges when the agent runs. Make sure you Set up consumption-based billing.

• Transcription is enabled for the channels that support voice conversations. For more information, see Enable transcription for voice channels.

• We recommend that you enable audit history and make sure service representatives have the required access to the case and related entities that the AI agent updates. Learn more in Manage Dataverse auditing.

• For customers to provide the details that the AI agent can use, you can configure preconversation survey.

• Make sure the service representatives working on the case and accepting conversations have read privileges on the msdyn_entityattributepredictionrules table.

• For the AI agent to predict case fields from emails autonomously, do the steps in Configure global settings for Case Management Agent.

Update field and lookup descriptions in Power Apps

To help the AI agent make better predictions for lookup fields, add descriptive information to your lookup records. Do the following steps in Power Apps:

• To help the AI agent understand the context, add meaningful field descriptions in your table columns. For example, in the Account Number column of the Account table, add a description like: "This is an account number. Account numbers start with ACC."
• Do the following steps to improve the AI agent's prediction accuracy with lookup fields:
  • For the required lookup entity, add a new optional text field to contain a description of the record if a description field doesn't already exist.
  • Add the meaning and usage for the description fields in the lookup records.
  • Update the Quick Find view of the lookup entity to include the new description field as a column.
  • Save and publish the changes.

For example, consider case categories like "Billing" and "Account Issues". When a customer writes "I can't access my account to pay my bill," it fits both categories. When you add clear descriptions to each lookup record, the AI agent can make more accurate predictions. If you include descriptions to the "Billing" category such as "Questions about charges and invoices, payment processing issues, refund requests," and "Login problems and password resets, profile updates and settings, account access difficulties" to "Account issues", the AI agent categorizes the customer's message as "Account Issues" because the primary problem relates to account access rather than billing.

Best practices for lookup descriptions

We recommend that you follow these guidelines when you add descriptions for lookup fields:

• Use simple, direct language and keep the descriptions under two or three sentences to ensure clarity. Don't add unnecessary information or domain jargon.
• Include typical scenarios, keywords, and phrases that users might use when describing their issue. This information helps the AI agent to make semantic connections between user input and the correct record.
• Distinguish between similar records by specifying what makes each record unique, preventing confusion and improving prediction accuracy.
• Don’t repeat the record name unless it adds clarity. Specify what the name doesn't convey about the record's intended use and scope.
• Use labeled sections like "Use when:" or "Not for:" to provide clear boundaries and usage guidelines for the AI agent.
• Provide synonyms, related terms, and specific examples. Avoid terms such as "general" or "miscellaneous" that lack meaningful context.
• Avoid overly generic descriptions, excessive detail, unexplained abbreviations, and assumptions about internal business logic that the AI agent can't access or understand.

Use Quick Find views to enable hierarchical lookup

Hierarchical lookup allows Case Management Agent to resolve values across related tables that are organized in a parent–child structure, such as categories and subcategories. When hierarchical resolution is used, the agent evaluates both the selected record and its related parent records to determine the most relevant match.

For these scenarios to work correctly, hierarchical resolution depends on how the Quick Find Active view is configured on the child table. The parent lookup column must be included in this view so that Case Management Agent can evaluate relationships and resolve hierarchical values during case creation and updates.

Prerequisites for hierarchical lookup

• A parent–child lookup relationship exists between the tables.
• The parent lookup column is available on the child table.
• You have permissions to edit, save, and publish views in Power Apps.

Example hierarchy

The following example shows a typical configuration for a hierarchical relationship:

Configure hierarchical lookup

The Quick Find Active view on the child table can include multiple lookup columns. To enable hierarchical lookup, make sure that the parent lookup column is included.

1. In Power Apps, go to Solutions, and then open the relevant solution.
2. Select the child table. For example, Product.
3. Select Views.
4. Open the Quick Find Active view, and then select View columns.
5. Find the parent lookup column (for example, Product family) and add it to the view.
6. Select Save and Publish.

Configure autonomous case updates

In Copilot Service admin center, configure the AI agent to predict and update case fields after a conversation ends or when processing an incoming email. The rules you specify in this section apply to all channels unless you explicitly configure them to apply to specific channels.

1. In Customer support, select Case settings.
2. On the Case settings page, select Manage for Case Management Agent.
3. On the Case Management Agent page, select Case creation and update with autonomous AI assistance.
4. In the page that appears, in Case update by AI agent (any channel), select Create. Specify the following information:
   • A unique name for the rule.
   • Conditions for the AI agent to apply the rule. If no conditions are defined, the rule applies to all channels.
   • Fields in Fields for AI prediction that the agent predicts and updates when the conversation ends or from an incoming email. If you don't specify update rules, the AI agent doesn't autonomously update any fields.
   • Select Save.

For example, if you only specify Issue description and Contact fields in the Fields for AI prediction section, the AI agent updates these fields when the conversation ends or from an incoming email. If you also specify a condition such as live chat status equals Active, then the rule applies only for live chat conversations that are active.

1. The system runs case update rules in the order they're listed. You can select the arrow buttons to reorder the rules as needed.
2. Select Activate to activate the rules.
3. Select Allow AI agent to override human edits during autonomous updates for the AI agent to automatically overwrite fields. During autonomous case update, the AI agent overwrites fields that service representatives previously edited.

Configure autonomous case creation

The following actions trigger the case creation process of Case Management Agent:

• The service representative accepts an incoming conversation request.
• The service representative ends the conversation.

To allow the AI agent to autonomously create cases across all provisioned messaging and voice channels, perform the following steps:

1. Go to Case creation and update > Case creation by AI agent (from chats and calls) and select Make Case Processing Agent available for case creation from conversations.
2. In Fields for AI prediction, specify the fields the agent predicts and populates in the case form using information from the conversation. The AI agent populates only those fields that have sufficient context available.

Run simulations to evaluate field prediction accuracy in Case Management Agent

Use simulation in Case Management Agent to validate the performance of AI‑powered field prediction on your organization’s historical data, sample email, or chat input. Assess prediction quality for confidence in the output before you enable the capability in production.

Set up a simulation

You can configure a simulation by using organization records or uploading an Excel file containing exported email or chat responses.

1. On the Case creation and update page, select Go to simulation in the Command menu. The Case creation and update simulation page appears.

2. On the Simulation setup tab, provide the following information:

   1. Simulation name: Provide a simulation name. For example: Surface product cases, Refund category evaluation, or Email‑based sample test.
   2. Data source: Select a data source from the dropdown.
      1. For Organization records, do as follows:
         1. Fields for AI prediction: Provide the fields for AI prediction.
         2. Conditions: Define the conditions to fetch the records to be used in simulation.
         3. Show records: Displays a list of records that you can select to use in simulation. You can select up to 100 case records.
      2. Excel Upload: Select an unencrypted file. The values are consolidated into a single string before they are used in prediction.
         1. Fields for AI prediction: Provide the fields for AI prediction.
         2. Upload File: Upload the simulation input Excel file. The maximum file size is 1 MB with a maximum of 100 records.

3. Select Run simulation.

Excel sample 1:

Excel sample 2:

View a simulation report

On the Case creation and update simulation page, go to the Simulation result tab. Simulations are listed with details of the simulation name, run date, status, result, average prediction match, and action.

• Select Download to export an Excel report. The report displays the record ID and predicted field values for each record.
• Select View. The Simulation overview page displays the following details:
  • Simulation setup (read-only): Configuration details used for the simulation.
  • Field prediction match: Details of the AI predictions when you select organization records. You can sort the list by prediction accuracy or alphabetically.
    • Prediction match (%) indicates how often predicted values match actual case values.
    • Cases are included only when both predicted and actual values are available. Text and multiline text fields are excluded.
  • Detailed view: Shows case-level details, such as actual and predicted values, and lets you add columns or download the data as an Excel file.
• Select Re-run if you modify field descriptions or prediction rules.

Best practices for data sources

• Start with small record sets (20–30 cases) to validate your field descriptions.
• Run multiple simulations focusing on different product lines, categories, and languages or regions.
• Refine field descriptions whenever prediction errors show recurring patterns.
• Test chats or emails with Excel files before launch.

Configure AI-assisted case creation for service representatives

Select the channels from which service representatives can create cases with AI assistance. You can select Email or Conversation (chats and calls). When a service representative creates a case from a conversation or an email, the AI agent analyzes the conversation or email, and then predicts and populates the fields available on the case form. Service representatives can then review the predicted values and make any necessary changes before saving the case.

Enable service representatives to use autonomous Case Management Agent

For service representatives to use Case Management Agent in Copilot Service workspace, allow the autonomous case creation and update, case follow-up, and closure in agent experience profiles.

By default, service representatives added to the out-of-the-box experience profiles can use the autonomous Case Management Agent.

1. Go to Experience profiles using one of the following navigation options:
   • Support experience > Workspaces
   • Select Manage for Case Management Agent, and then select agent experience profiles in Case creation and update > Representative access.
2. Select the required experience profile.
3. In the Copilot AI features section, do the following actions:
   • Select From conversations in Autonomous case creation and update.
   • In Form fill assistance for cases select During case creation from conversation and During case creation from email to indicate which channels the AI agent can assist service representatives in creating cases.

Record representative interactions with the AI agent

In Agent experience data from Representative experience data, you can select Record transcripts of representative interactions with AI, including representative actions, and their feedback on AI suggestions to record and understand how representatives are interacting with the AI agent and how the agent is performing in a support organization. Representatives can also share feedback about AI agent actions, which helps Copilot perform better. You can also download and use the data to analyze knowledge sources, and build usage reports.

Example

When a customer initiates a chat conversation with the service representative, the AI agent creates a case if there's enough context to update at least one of the Issue description or Contact fields.

For the agent to run this scenario, specify the following in the Case creation and update page:

• Channel: Chat
• Fields for AI prediction: Issue description, Contact

When the conversation ends, the AI agent must update the Issue description and Contact fields, if there are any updates. The Product, Priority, and Serial number fields should also be updated if the case category is set to product defect.

For the agent to run this scenario, in addition to the Issue description and Contact fields set in Fields for AI prediction, specify the following in the Case update by AI agent (any channel) section:

• Select Create for Case update rules.
• In the New rule page, specify the Rule name and the following:
  • In Define conditions, select Add and then specify the following:
    • Select a field: Case category
    • Operator: Equals
    • Value: Product defect
  • Specify fields for AI prediction when this condition is met:
    • Product, Priority, Serial number

Next steps

Use Case Management Agent to create and update cases